Method for geometric element segmentation of a point cloud of measurements of a single complex component

Abstract

The method for segmenting geometric elements of a single complex component measurement point cloud is characterized in that: after obtaining the point cloud data, the region growing method is used to divide independent regions, the independent regions representing boundaries are recorded as boundary regions, and other independent regions are respectively recorded as geometric feature regions; and the points in the boundary regions are distributed to the geometric feature regions. The method has the advantages that: when the geometric element segmentation of the single complex component measurement point cloud is performed, the geometric deviation is considered, all points can be distributed to the independent regions without losing information, and the geometric element segmentation of the product containing the geometric deviation can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to a point cloud segmentation method, in particular to a method for segmenting geometric elements of a single complex measurement point cloud; the method is suitable for geometric element segmentation before quality evaluation of a single complex component in the fields of aviation, nuclear power equipment, etc. BACKGROUND

[0002] Point cloud is a collection of object surface point data, and is a commonly used model representation method. Point cloud contains information of all surfaces of a model, and has a wide range of applications in error analysis, free-form surface reconstruction, machine vision, etc. When performing related analysis of a model, a specific surface of the model is targeted. Therefore, when analyzing, point cloud subsets belonging to different patches need to be segmented from the point cloud. The essence of point cloud segmentation is to divide points corresponding to different features into the same point set according to the feature information of the model.

[0003] At present, point cloud segmentation methods can be divided into three categories according to different processing methods: edge-based methods, face-based methods and clustering-based methods. However, most edge-based methods are not only affected by measurement noise, but also not easy to find the boundaries of surfaces with little curvature change. When using face-based methods, the edges of the point cloud are prone to deformation, and the problems of "over-segmentation" and "under-segmentation" may occur. When using clustering-based methods, it may be difficult to find the basis for clustering. At the same time, a large amount of neighborhood information of points needs to be calculated during clustering. Therefore, clustering-based methods are prone to incorrect segmentation of sparse point clouds or point clouds with uneven density.

[0004] The above methods extract feature information of a model based on local information of a point cloud, and thus the segmentation of the point cloud is based on local information of the point cloud. Therefore, the above methods have poor accuracy.

[0005] Chinese patent CN107784656B proposes a part point cloud segmentation method based on geometric elements, which uses IGES model and part point cloud registration involving software to realize geometric element segmentation of part point cloud. The problem of this segmentation method is that it cannot be applied to components without IGES model. SUMMARY

[0006] The present application is directed to geometric operations defined by ISO GPS, including specification and verification fields. This operation is used to identify geometric features of interest from the entire component (workpiece), and the object may be a specification model or measured data obtained by evaluating measurement.

[0007] In actual engineering, the division of geometric features is carried out before the quality evaluation of products. The measured data is the point cloud data of products containing geometric deviations. Therefore, when the geometric features are divided, the geometric deviation factor should be considered. However, if the geometric deviation is introduced, the direction of the normal vector of each point will be affected. For example, if the classic region growing method is used to segment the cube point cloud data, the cube will be divided into seven independent regions, one of which is the edge of the cube. The region growing method relies on the curvature value of each point in the model. The curvature mutation is an important reference for segmentation. However, the geometric deviation makes the boundary of the cube no longer clear, but becomes a small free-form surface. Therefore, some growing regions will cross the boundary at the place where the curvature changes slowly and grow to another surface.

[0008] Cai et al. proposed a hybrid segmentation process and Qie et al. proposed an enhanced mesh segmentation method using curvature and shape index (two shape-based surface descriptors) to preliminarily divide the point cloud of geometric products. By comparing the values of these two descriptors of each vertex, the entire model is divided into ten shape types and sharp edges. When the boundary is sharp (thin) to the extreme, the independent regions are divided with clear boundaries, and then the boundary is ignored to complete the extraction of geometric features. This method has the following problems in geometric feature segmentation: 1. The points (or triangular meshes) on the boundary are not segmented into other independent regions, and the lost point cloud data may affect the subsequent quality evaluation of the components (such as tolerance, part (component) deformation, etc.); 2. The sharper the boundary, the more likely it is to appear boundary fracture, which will hinder the identification of independent regions.

[0009] The purpose of the present application is to provide a method for segmenting geometric elements of measured point cloud of a single complex component, which can divide all points into independent regions.

[0010] The method for segmenting geometric elements of measured point cloud of a single complex component, characterized in that: after obtaining the point cloud data, the independent regions are divided by using the region growing method, and the independent regions representing the boundary are recorded as boundary regions, and the other independent regions are recorded as geometric feature regions respectively.

[0011] The points in the boundary region are assigned to the geometric feature regions, which needs to satisfy the following two conditions: 1, the geometric shape of the current geometric feature region does not change significantly after the points in the boundary are assigned to the current geometric feature region; 2, a point in the boundary is surrounded by adjacent points, if the point and the adjacent points surrounding it are divided into different regions, the division of the point is considered as interference, and the point is re-divided into the region of the adjacent points; 3, two or more points in the boundary are collinear, if the line segment and the adjacent points surrounding it are divided into different regions, the division of the line segment is considered as interference, and the points on the line segment are re-divided into the region of the adjacent points; after all the points in the boundary region are assigned to the geometric feature regions, the geometric element segmentation is completed.

[0012] Further, the assignment of the points in the boundary region to the geometric feature regions includes the following steps:

[0013] Step 1: obtaining two adjacent geometric feature regions intersecting the current boundary, obtaining the current two adjacent geometric feature regions and performing plane fitting respectively to obtain plane A and plane B; the plane fitting equation can be: px+qy+mz+Q=0; (x, y, z) is the coordinate of the point, p, q, m represent the coefficients, and Q is the constant;

[0014] Step 2: for each point n j (x j ,y j ,z j ) in the boundary region, according to the equation

[0015]

[0016] , the distances of the points to the planes A and B are calculated and

[0017] Step 3: for each point n j (x j ,y j ,z j ), the values of and are compared; the point n j (x j ,y j ,z j ) is added to the geometric feature region with smaller distance, and the current geometric feature region is updated;

[0018] Step 4: checking the divided points n j (x j ,y j ,z j ):

[0019] Case I: If a point belongs to the plane and all the points connected to it belong to the B region, then the attribution of the vertex is transferred to the B region; if a point belongs to the plane and all the points connected to it belong to the A region, or only one connected vertex belongs to the B region, then the attribution of the vertex is transferred to the A region;

[0020] Case ii: A point belongs to the A region, and the A region points connected to the point are not more than two, the point and the points connected to the point in the A region are taken as a set of points to be screened, and a point in the set of points to be screened is taken as a current point to be judged, and all the points connected to the current point except the set of points to be screened belong to the B region, then the attribution of the set of points to be screened is transferred to the B region;

[0021] A point belongs to the B region, and the B region points connected to the point are not more than two, the point and the points connected to the point in the A region are taken as a set of points to be screened, and a point in the set of points to be screened is taken as a current point to be judged, and all the points connected to the current point except the set of points to be screened belong to the A region, then the attribution of the set of points to be screened is transferred to the A region;

[0022] Step 5: Judge whether all the points in the boundary region have been calculated, if yes, complete the geometric feature segmentation.

[0023] Further, step 5 is modified as: setting an optimization iteration termination condition Where N b is the number of points contained in the boundary that need to be refined, b m is the number of points whose attribution region changes after the mth iteration, and ε is a default value, and ε << 1;

[0024] If it is satisfied, the iteration is ended; otherwise, the updated geometric feature region is taken as the current geometric feature region, and steps 1-5 are repeatedly executed. This termination condition means that after several iterations, the division of most points in the original boundary no longer changes. The smaller ε is set, the higher the accuracy of the division is.

[0025] Further, when the region boundary method is used to segment the point cloud, the point cloud to be segmented is denoted as n, n i is an arbitrary point in the point cloud, and the point set satisfying E(n) = {n i | c(n i )-c min > c e} is regarded as a boundary region point set, wherein, K1(n i ) and K2(n i ) represent two principal curvatures of the n i point; C min is the minimum value of C(n i ), Cmax is the maximum value of C(n i ); δ is a coefficient, and its value ranges from 0.05 to 1.

[0026] Further, the value of δ ranges from 0.05 to 0.15. This value range makes the boundary region not to be broken, and intentionally increases the inaccuracy of the points in the boundary region.

[0027] The present application has the advantages that, when the geometric element segmentation of the measurement point cloud of a single complex component is performed, the geometric deviation is considered, and all points can be divided into independent regions without losing information; and the geometric element segmentation of a product containing geometric deviation can be realized. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 is a schematic diagram for checking the divided points n j (x j , y j , z j ). DETAILED DESCRIPTION

[0029] The method for performing geometric element segmentation on the measurement point cloud of a single complex component comprises the following steps: after the point cloud data is acquired, the independent regions are divided by using the region growing method, the independent regions representing the boundaries are recorded as boundary regions, and the other independent regions are respectively recorded as geometric feature regions;

[0030] The points in the boundary region are assigned to the geometric feature regions, and the following two conditions need to be met simultaneously: 1. After the points in the boundary are assigned to the current geometric feature region, the geometric shape of the current geometric feature region does not change significantly; 2. A certain point in the boundary is surrounded by adjacent points, if the point and the adjacent points surrounding it are divided into different regions, the division of the point is regarded as interference, and the point is re-divided into the region of the adjacent points; 3. A certain two or several points in the boundary are collinear, if the line segment and the adjacent points surrounding it are divided into different regions, the division of the line segment is regarded as interference, and the points on the line segment are re-divided into the region of the adjacent points; after all the points in the boundary region are assigned to the geometric feature regions in the condition, the geometric element segmentation is completed.

[0031] In order to prevent the identified boundary from being broken and possibly leading to the uncontrolled expansion of the region outside the boundary region, the points close to the sharp edge are identified as part of the sharp edge, thereby leading to a rough boundary. Taking a cuboid as an example, the six planes obtained by division are incomplete. It is necessary to refine them by dividing the rough boundary into geometric elements connected thereto. The sampling data is a grid by default.

[0032] In some embodiments, the assigning of the points in the boundary region to the geometric feature regions comprises the following steps:

[0033] Step 1: Get two adjacent feature regions intersecting the current boundary, get the current two adjacent feature regions and respectively perform plane fitting to obtain plane A and plane B; the plane fitting equation can be: px + qy + mz + Q = 0; (x, y, z) is the coordinate of the point, p, q, m represent the coefficient, and Q is the constant;

[0034] Step 2: For each point n j (x j ,y j ,z j ) of the boundary region, calculate their distances to plane A and B according to the equation

[0035]

[0036] and and

[0037] Step 3: For each point n j (x j ,y j ,z j ), compare the values of and ; add point n j (x j ,y j ,z j ) to the feature region with smaller distance, and update the current feature region;

[0038] Step 4: Check the situation of the divided point n j (x j ,y j ,z j ), as shown in Figure 1 :

[0039] Case I: If a point belongs to the plane, and all the points connected to it belong to the B region, then the attribution of the vertex is transferred to the B region; if a point belongs to the plane, and all the points connected to it belong to the A region, or only one connected vertex belongs to the B region, then the attribution of the vertex is transferred to the A region;

[0040] Case ii: A point belongs to the A region, and the A region points connected to the point do not exceed two, the point and the points in the A region connected to the point are taken as the screening point set, and a point in the screening point set is taken as the current point to judge; all the points connected to the current point except the screening point set belong to the B region, then the attribution of the screening point set is transferred to the B region;

[0041] If a point belongs to the B region and the number of points connected to the point in the B region is no more than two, the point and the points connected to the point in the A region are taken as a point set to be screened, and a point in the point set to be screened is taken as a current point. If all the points connected to the current point belong to the A region except the point set to be screened, the attribution of the point set to be screened is transferred to the A region.

[0042] Step 5: judging whether all the points in the boundary region have been calculated, if yes, the geometric feature segmentation is completed.

[0043] In some embodiments, step 5 is modified as: setting an optimization iteration termination condition wherein N b is the number of points contained in the boundary to be refined, b m is the number of points whose attribution changes after the mth iteration, and ε is a default value and ε << 1.

[0044] If the termination condition is met, the iteration is ended; otherwise, the updated geometric feature region is taken as a current geometric feature region, and steps 1-5 are repeatedly executed. The termination condition means that after several iterations, the division of most points in the original boundary no longer changes. The smaller ε is set, the higher the division accuracy is.

[0045] In some embodiments, when the region boundary method is used to segment the point cloud, the point cloud to be segmented is denoted as n, n i is any point in the point cloud, and the point set satisfying E(n) = {n i | c(n i ) - c min > c e} is taken as a boundary region point set, wherein, K1(n i ) and K2(n i ) represent two principal curvatures of the point n i .

[0046] C min is the minimum value of C(n i ), C max is the maximum value of C(n i ), and δ is a coefficient with a value range of 0.05-1.

[0047] In some embodiments, the value of δ is 0.05-0.15. The value range makes the boundary region not appear to be broken, and intentionally increases the inaccuracy of the points in the boundary region.

[0048] The above specific implementation cases are used to explain and illustrate the application, and are only preferred embodiments of the application, but not limit the application, and any modification, equivalent replacement, improvement, etc. made to the application falls within the protection scope of the application.

Claims

1. Method for geometric feature segmentation of a point cloud of measurements of a single complex component, characterized in that: After obtaining the point cloud data, independent regions are divided by using a region growing method, and the independent region representing the boundary is recorded as a boundary region, and the other independent regions are recorded as geometric feature regions respectively; The points in the boundary region are assigned to the geometric feature regions, which need to meet the following three conditions:

1. After the points in the boundary are assigned to the current geometric feature region, the geometric shape of the current geometric feature region does not change significantly; 2. A point in the boundary is surrounded by adjacent points, and if the point and the adjacent points surrounding it are divided into different regions, the division of the point is regarded as interference, and the point is re-divided into the region where the adjacent points are located; 3. Two or more points in the boundary are collinear, and if the line segment formed by the collinear points and the adjacent points surrounding it are divided into different regions, the division of the line segment is regarded as interference, and the points on the line segment are re-divided into the region where the adjacent points are located; after all the points in the boundary region are assigned to the geometric feature regions, the geometric feature segmentation is completed. The assignment of the points in the boundary region to the geometric feature regions includes the following steps:

2. The method of geometric element segmentation of a measurement point cloud of a single complex component according to claim 1, characterized in that: Case I: If a point belongs to plane A and all the connected points belong to region B, then the attribution of the point is transferred to region B; if a point belongs to plane B and all the connected points belong to region A, or only one connected vertex belongs to region B, then the attribution of the vertex is transferred to region A; Step 1: Obtain two adjacent geometric feature regions intersecting the current boundary, obtain the current two adjacent geometric feature regions and respectively perform plane fitting to obtain plane A and plane B; the plane fitting equation can be: ; (x, y, z) is the coordinate of a point, p, q, m represent the coefficients, and Q is a constant; Step 2: For each point n of the boundary region j (x j ,y j ,z j ), according to the equation Calculate their distance to the planes A and B and ; Step 3: for each point n j (x j ,y j ,z j ), compare and values; add point n j (x j ,y j ,z j ) to the region of the geometric feature with smaller distance, update the current geometric feature region; Step 4: Check the point connection condition of the divided n j (x j ,y j ,z j ) Case II: A point belongs to region A, and the number of A-region points connected to the point is not more than two, and the point and the points connected to the point in region A are taken as a set of points to be screened, and a point in the set of points to be screened is taken as a current point to be judged, and all the points connected to the current point except the set of points to be screened belong to region B, then the attribution of the set of points to be screened is transferred to region B; A point belongs to region B, and the number of B-region points connected to the point is not more than two, and the point and the points connected to the point in region B are taken as a set of points to be screened, and a point in the set of points to be screened is taken as a current point to be judged, and all the points connected to the current point except the set of points to be screened belong to region A, then the attribution of the set of points to be screened is transferred to region A; Step 5: Determine whether all the points in the boundary region have been calculated, if yes, the geometric feature segmentation is completed. If yes, the iteration is ended; otherwise, the updated geometric feature region is taken as the current geometric feature region, and steps 1-5 are repeatedly executed; 3. The method of geometric element segmentation of a measurement point cloud of a single complex component according to claim 2, characterized in that: Step 5 is modified as: Set the optimization iteration termination condition ; where N b is the number of points contained in the boundary that needs to be refined, b m is the number of points whose attribution region changes after the mth iteration, and ε is a default value and ε <<1; This termination condition means that after several iterations, the division of most points in the original boundary no longer changes. The value of δ is 0.05-0.15, which makes the boundary region not appear to be broken, and intentionally increases the inaccuracy of the points in the boundary region.

4. The method of geometric element segmentation of a measurement point cloud of a single complex component as claimed in claim 1, characterized in that: When the point cloud is segmented using the region boundary method, the point cloud to be segmented is denoted as n, n i As an arbitrary point in the point cloud, the point set satisfying is regarded as a boundary region point set, wherein , K1(n i ) and K2(n i ) represent two principal curvatures of the point n i ; , C min is the minimum value of C(n i ), C max is the maximum value of C(n i ); and δ is a coefficient, and the value range of δ is 0.05-1.

5. The method of geometric element segmentation of a measurement point cloud of a single complex component as claimed in claim 4, characterized in that: ​

Images